That viral photo of the Pope in a puffer jacket. The “arrest” photos that fooled newsrooms. AI-generated images are now good enough to deceive at a glance — which means knowing how to detect AI-generated images vs real photos is becoming a basic literacy skill. I’ve studied thousands of AI images and the tells that give them away. Here’s your field guide: 9 signs to check, plus the tools that help.
The 9 Telltale Signs of AI-Generated Images
1. Hands and Fingers
The classic tell. Count the fingers. Look for fused digits, extra joints, hands melting into objects. AI has improved enormously here, but hands in complex poses still glitch regularly. Always check hands first — it takes two seconds.
2. Text and Letters
Signs, labels, book spines, t-shirt text — AI mangles text more than anything else. Gibberish lettering, misspelled words, letters that morph mid-word. Newer models are better (see our text-in-images guide), but background text still fails often.
3. Too-Perfect Skin
AI faces often have an airbrushed, waxy quality — impossibly smooth skin with no pores, no blemishes, no asymmetry. Real human skin has texture. The “beauty filter look” on every face in a crowd photo is a red flag.
4. Inconsistent Lighting and Shadows
Shadows falling in different directions. Light sources that don’t match. Earrings where one reflects a window and the other doesn’t. AI composes scenes statistically, not physically — physics errors are common.
5. Background Weirdness
Zoom into the background. Blurred crowds with melted faces, architecture that doesn’t quite connect, objects merging into each other. AI spends its detail budget on the subject; backgrounds get mushy and surreal.
6. Repetitive Patterns and Textures
Brick walls with impossible geometry, fabric patterns that don’t repeat correctly, hair that’s too uniform. AI struggles with precise repetition.
7. Ears, Teeth, and Glasses
The supporting cast of tells: misshapen ears (especially with long hair), too many teeth or teeth merging together, glasses with asymmetric frames or impossible reflections.
8. The “Vibe” — Overly Polished Unreality
Harder to define but real: AI images often have a hyper-polished, slightly dreamlike quality — perfect composition, dramatic lighting, everything a bit too ideal. Real photos have awkwardness: bad framing, weird expressions, messy backgrounds.
9. Metadata and Provenance
The technical check: AI generators increasingly embed C2PA metadata identifying images as AI-created. Tools and browser extensions can read this. Absence of metadata proves nothing (it can be stripped), but presence is conclusive.
Detection Tools
- Hive Moderation / Illuminarty: dedicated AI image detectors — decent, not definitive. They give probabilities, not verdicts.
- Google reverse image search: if the “photo” appears only in AI galleries or nowhere, that’s informative.
- C2PA verification tools: check embedded provenance metadata.
- Your eyes: honestly the best tool. Train yourself on the 9 signs above.
Important caveat: detectors have false positives and false negatives, just like text AI detectors. Use them as signals, not proof.
Why This Matters
This isn’t academic. AI-generated images are used in scams (fake disaster photos for donations), misinformation (fake news events), fraud (fake product photos, fake identities), and social engineering. The laws around AI deepfakes are still catching up — which means personal vigilance is the main defense right now.
And if you’re a creator: disclose AI use in your content. Transparency builds trust; getting caught doesn’t.
How to detect AI-generated images vs real photos: a field guide
Hands remain the classic tell, though they’re improving. Look for: wrong finger counts, fused or melting fingers, hands that blend into objects, watches or jewelry that morph impossibly. Zoom in — thumbnails hide what full resolution reveals.
Beyond hands: text in the image (signs, labels — AI still garbles these), background coherence (crowd faces that dissolve, architecture that doesn’t quite connect), lighting consistency (shadows pointing different directions), and too-perfect skin (that waxy, poreless look).
No single tell is conclusive — real photos can have odd artifacts, and the best AI images pass casual inspection. Combine multiple signals, check provenance (reverse image search), and consider the source’s credibility. For understanding the generation side, our guide on checkpoint models for realistic photos explains what these tools can do, and Midjourney’s documentation covers the technology.
Video is next: AI-generated video still shows tells — inconsistent physics, morphing objects, temporal flicker — but they are subtler than image tells and improving fast. The detection skills transfer: look for coherence across frames the way you check coherence across an image. Expect this to be a permanent arms race.
The Bottom Line
How to detect AI-generated images vs real photos: check hands, text, skin texture, lighting consistency, and backgrounds — the 9 signs above catch most fakes in under a minute. Use detection tools as backup, not gospel. And cultivate the habit: before sharing a surprising image, spend 30 seconds looking closely. In 2026, that’s not paranoia — it’s basic media literacy.
